SOFTWARE CORNER Blade Servers: Evolution and revolution

By Curtis A. Schwaderer

Blade servers are a revolutionary new concept for enterprise replacing older, slower blades with new blades that have higher applications currently using a “stack of PC servers” approach. performing or symmetric multiprocessor blades. Many blade At the same time, blade servers promise to be evolutionary in solutions are “scale-out” because of lower cost of owner- the communications market where using I/O blades in a network ship, but some vendors have scale out and scale up strategies. system is the norm. Blade servers promise to greatly increase compute density, reduce cost, improve reliability, and simplify Blade servers also lower capital expense. Blade server technology cabling. Companies such as , Hewlett Packard, IBM, RLX, leverages high volume standard technologies like processors, and Sun offer blade server solutions that reduce operating memory, operating systems, , and the Internet Protocol expense while increasing services density. (IP) in a form factor that significantly lowers system expense. Blade Servers generally incorporate Intel Architecture (IA), In this article, I’ll be covering blade server vendors’strategies SPARC, and Transmeta processor technologies along with and solutions from a unique perspective Ð the three pillars of HP-UX, Linux, Solaris, and Windows operating systems. By a balanced end-to-end system Ð compute power, data store, and leveraging volume generated by other markets, the blade server I/O. A very big thanks goes out to Dell, Hewlett Packard, IBM, bill of materials cost is significantly lower. Scalability is another Intel, RLX, and Sun for their participation and cooperation in major blade server advantage. Chassis can be purchased with just putting this information together. enough compute blades to accommodate today’s capacity require- ments. Additional or higher performance compute blades can be Compute power, data store, and I/O populated as needed over time, spreading the capital expense Depending on where you sit in the end-to-end system solution, the when the provider knows it will be profitable to do so. Most blade proper balance of compute power, data store, and network I/O is server vendors also enable on-line system scaling, so capacity can critical to optimal performance. The endpoints of the end-to-end be increased without down-time. system tend to be compute power and data store heavy. Your PC needs to have the latest processor with gigs of hard drive space to There are also differences between the enterprise and communica- play that latest interactive network game. At the other end of the tions target markets. The enterprise market wants to create a com- system, there is a server managing the virtual world for that inter- pute power and data store heavy blade server system that sits at or active game that must also be compute and data store heavy. As near the network endpoints. The communications market wants to you push from the endpoints to the middle of the network, net- have increased communications and packet processing blade work traffic and I/O demands dominate. Inside the network itself, options for I/O and sprinkle in enough compute power and storage I/O is critical, but billing systems and advanced intelligent net- to perform billing and advanced intelligent network (AIN) services. work features require an increase in the compute power part of the equation. Blade servers are a great low-cost way to add compute General blade server architecture power to network infrastructure in addition to traditional enter- A general blade server architecture is shown in Figure 1. The prise solutions. hardware components of a blade server are the switch blade, chas- sis (with fans, temperature sensors, etc), and multiple compute Market need for blade servers blades. Some vendors offer, partner, or plan to partner with com- The market for blade servers can be divided into enterprise systems panies that provide application specific blades that provide traffic (data centers, internet service provider server farms) and communi- conditioning, protection, or network processing prior to the traffic cations systems (wireless, wire-line, packet cable, broadcast, and reaching the compute blades. Often, these application specific the Internet). Service providers in enterprise and communications blades may be functionally positioned between the switch blade are both looking for more highly capable, high-availability solu- and compute blades. However, these blades reside in a standard tions that also decrease overall capital and operational expense. compute blade slot.

Blade server technology provides a significant density advantage The outside world connects through the rear of the chassis to a to traditional enterprise systems. The stacked PCs enterprise sys- switch card in the blade server. The switch card is provisioned to tems tend to take up a large amount of space. Blade servers put the distribute packets to blades within the blade server. All these com- compute platform functionality on a circuit board with specially ponents are wrapped together with network management system designed chassis that integrate management and cooling to accom- software provided by the blade server vendor. The specifics on the modate a large number of blade slots within a single chassis. This blade server architecture vary from vendor to vendor. But before results in far more compute power per area than the stacked PC you discount this as a bunch of proprietary architectures, think approach. Adding more blades to the system is what the industry again. Remember that IBM and others dramatically advanced and calls “scaling out.” In other words, add more blades as your com- proliferated the PC architecture, changing the face of computing pute power requirements increase. A “scale up” solution implies forever. The blade server industry appears to be headed in the same

Reprinted from CompactPCI Systems / March 2003 Figure 1 direction. There are some areas where standardization of blade dant power/cooling and management functionality to help lower server components will prove helpful. However, blade server ven- cost, reduce cables, and simplify servicing and maintenance. dors ability to quickly adapt and advance their architectures to suite specific applications unencumbered by the standards process Blades available for the 1655MC include Intel Pentium III proces- will prove to accelerate proliferation in the near term. sors operating up to 1.4 GHz. Two hot plug redundant 1048 watt power supplies with 100-240 volt AC support powers the system. Compute blades and application specific blades The 1655MC supports up to two managed layer-2 Ethernet The blade server purist will tell you that there is a compute blade switches. Each switch has six Gigabit Ethernet (GbE) inputs with solution for every problem. For example, if load balancing or four copper GbE uplinks. The uplink ports support trunking and security is required, the compute blades will have load balancing port aggregation. Each server blade includes two embedded and security software on them. In essence, you take the same Gigabit network interface cards (NICs). generic compute blade, and load the software necessary to solve the problem at hand. Communications-oriented blade server tech- The 1655MC consists of a 3U enclosure that holds up to six server nologists assert application specific blades (i.e. blades with blades. A standard 42U rack holds 14 enclosures for a total of 84 unique processing such as a network processor or ASIC designed server blades. Each server blade supports up to two Intel PIII specifically to accelerate packet processing within the blade processors, two 73 Gbyte SCSI hard drives with integrated RAID, server architecture) are required. and 2 Gbyte ECC SDRAM.

Application specific blades offload processing that would otherwise Future blade offerings will provide greater flexibility in process- be hosted on every compute blade into one central, redundant loca- ing, storage, and I/O for larger deployments. Customers will be tion. The switch module in Figure 1 is the most basic example of an able to mix and match storage blades, processor blades, and I/O application specific blade. Other application specific blades that blades according to their computing needs. deal with load balancing, security, attack detection and prevention, and authorization. Blades that may have specialized silicon such as One unique aspect of Dell's strategy is to provide their customers a network processor or ASIC and software optimized to process with low cross-over pricing when they are ready to migrate traffic flows from layer 3-7 at speeds up to Gigabit Ethernet and toward a blade architecture. For example, if more than three beyond. Application specific blades can lower compute blade cost existing dual processor blade, 1U rack servers are required, it will while also increasing overall blade server performance and capac- cost less to migrate to the 1655MC server blades. Dell wants to ity. In short, application specific blades like network processing protect their existing customer base, so their server blade systems blades balance out the I/O dimension of the compute, data store, are designed to co-exist with traditional rack-optimized servers, and I/O equation for more network-centric deployments. without requiring modification to the customers’ rack or power environment. Who’s who in blade server vendors DELL As part of their Intelligent IT portfolio, Dell ships their Open- Dell’s blade server strategy is to bring systems to market that Manage Remote Install tools with the 1655MC for no extra improve rack density and reduce cabling requirements while charge. This enables customers to provision hundreds of blades being able to seamlessly integrate with and transition from tradi- over the network from a centralized deployment station. tional rack-optimized server environments. OpenManage tools also allow customers to remotely manage, monitor status, and control the enclosure and individual server Dell blade server target applications include Web serving, thin blades. Users can also configure individual blades locally client computing, high performance clustering, application serv- through a bootable USB floppy or CD-ROM drive. Dell is also ing, security, and network infrastructure applications. The making it easy for its customers to continue to use their current PowerEdge 1655MC was announced in April 2002 and began imaging and provisioning tools to deploy Dell server blades. This shipping in November 2002. It’s designed to enable customers to are examples of Dell's low cross-over pricing strategy and are improve rack density by about 50 percent while reducing cable also important for seamless network management and provision- requirements by as much as 80 percent. The 1655MC server ing across the entire enterprise. blades provide integrated server management and software deployment, share a common 3U form factor, and leverage redun- Dell partners with Microsoft and Red Hat to deliver Windows and

Reprinted from CompactPCI Systems / March 2003 SOFTWARE CORNER

Linux server platforms. Applications partners include Citrix for visioning can be set by time or event to add and remove applica- thin-client computing and FP for network infracture/Web solutions. tions from blades.

HP Insight Manager 7 is the system level management tool that HP focuses their blade server solutions on investment protection, enables comprehensive fault monitoring and system software providing a complete portfolio of blade servers, heterogeneous maintenance for ProLiant servers. Within a blade server infra- SAN connectivity, flexible interconnect options, and built-in structure, it allows the administrator to pull up a visualization of power and cooling headroom for accommodating multiple gener- the rack, see the status of the blade, update the system software, ations of 1, 2, and 4 processor blades. and launch a remote console session on the blade for full admin- istrative control of the server. These system management and pro- Prior to the HP/ merger, both companies offered blade visioning tools hook into the HP OpenView environment for a server products. The Compaq ProLiant BL line of blade servers comprehensive service level management solution. were focused on enterprise applications and Web hosting, while the HP Powerbar line was geared toward communications. After HP has a partners program. In addition to operating system and the merger, consolidation began. It became difficult to put application software vendors, they also have some unique part- together one roadmap that satisfied a critical mass within the com- nerships with other companies. On the enterprise side, HP has munications market. On the communications side, Christine partnered with Altiris in delivering the ProLiant Essentials Rapid Martino, director of carrier grade servers, said HP will be address- Deployment Pack and has partnered with other provisioning ing the needs of and network service partners like Jareva to offer customers flexibility to choose their providers with an integration and services approach to better meet provisioning tools. their customized needs. This allows HP to provide these cus- tomers with an increased variety of application specific I/O and Hewlett Packard has an OpenCall business unit that provides sys- compute blades via HP and third-party partner products. tem integration services for customers. They have a large suite of software products available that can be integrated with blade The ProLiant BL line continues toward extending the ProLiant server hardware for things like SS7, management, and business product line as the corporate data center and service provider mar- applications. For more information, visit www.hp.com/servers/ kets begin more rapid adoption of blade server technology. blades

The next-generation ProLiant BL20p (p-Class) server uses dual IBM processor Xeon technology, each running up to 2.8 GHz. The new IBM blade server solutions focus on high performance, reliability, ProLiant BL40p server uses Xeon MP processor technology run- and ease of use and are targeted at enterprise-level large data cen- ning up to four 2.0 GHz processors in a highly available blade ter applications. form factor. HP is the first company to offer 1, 2, and 4 processor blades, making the solution attractive for multi-tiered enterprise When developing the blade, IBM highlighted seven important applications. aspects to blade servers which they dubbed “The magnificent seven”: HP also offers a lower end ProLiant BL e-Class blade server prod- uct. The e-Class enclosures house 20 3U blades and 280 servers in a ■ High density full-size 42U rack. The ProLiant p-Class and e-Class servers are ■ Reduced power backward compatible with existing BL enclosures. HP has opti- ■ Reduced cost mized the next-generation ProLiant BL20p and BL40p blade servers ■ Performance for HP StorageWorks, but provides heterogeneous SAN storage sup- ■ Reliability/availability port, so a variety of SAN options work with the ProLiant products. ■ Systems management ■ Ease of use Operating system support for the blade servers includes Linux and Windows. Management solutions for the ProLiant systems At the same time, IBM chose to place special emphasis on per- include HP Insight Manager 7 and HP ProLiant Essentials Rapid formance, reliability, and ease of use. Deployment Pack (RDP), and the advanced integrated lights-out technology embedded in the p-Class blades. Prior to IBM eServer BladeCenter that was announced in April 2002, blades were thought to be mainly a telecom play. IBM The ProLiant Essentials Rapid Deployment Pack product is a changed the game by bringing enterprise-level computing and provisioning tool. It allows you to pre-determine each blade’s reliability to its blade offering. personality before the blades are deployed. When blades are inserted, RDP will auto-provision the OS and application images IBM's chassis supports up to 14 blades, in a highly-available on each blade in the system. This capability speeds installation, infrastructure. The switch devices are broken into two halves Ð ensures consistency in provisioning, and simplifies management this way you can have two of the four switches be Ethernet and the of large numbers of blades. On blade failure, a new blade can be remaining two be , I/O media, or other interconnect. inserted and RDP takes care of auto-provisioning the server with The modular switch approach enables support for a wider variety the correct image for the slot. RDP is also used for re-provision- of I/O beyond Ethernet. In addition, IBM plans to provide future ing. New images can be dragged and dropped onto the blade icon switch modules capable of handling Ethernet layer 4-7 switching in the management system console and the designated server within the chassis. or set of servers will be automatically re-provisioned. Repro-

Reprinted from CompactPCI Systems / March 2003 SOFTWARE CORNER

IBM’s blades emphasize performance through their capacity for management locally or remotely across the entire system. two 2.4 GHz Intel Xeon processors. While this may present power ■ Extensive provisioning aspects. An administrator can create a and heat dissipation challenges on the chassis design, IBM enclo- SW image and deploy it rapidly across hundreds of blades. For sures use calibrated vector cooling to ensure the system can example, one RLX customer in the technical computing area accommodate the power/thermal issues. It will also allow the deployed over seven hundred blades across 42 racks. Control blade to accommodate larger, more powerful processors in the Tower was used to create and deploy the software image over future providing investment protection for customers. Ethernet from a remote location. This provisioning took only 20 minutes, a significant time savings compared with a manual Windows and Linux operating systems are supported image-loading approach. Future extensions include automatic in the BladeCenter. re-provisioning an existing blade to take on a different task The chassis, midplane, and switch module architecture of the dynamically. The dynamic nature is envisioned to go to the BladeCenter provides flexible compute, I/O, and storage options extent of allowing server re-provisioning that might be auto- that allows IBM to play in both the enterprise and telecom space. matically performed through set trigger events set within The dimensions are somewhat different, conforming to telecom Control Tower. Control Tower also has the ability to configure requirements such as 20-inch depth and NEBS compliance. the blade server system in redundant/spare fail-over configu- rations. In this case, you may have generic compute blades IBM sponsors a BladeCenter Alliance Partners program with a designated as stand-by. When blade failure is detected, Control charter to provide a variety of hardware and software options, Tower can load a copy of the software image of the failed blade evolve standards, and unify management interoperable with other and re-provision for lower costs associated with high-avail- devices developed within the partners program. IBM's enterprise ability configurations. focus caters to software application partners like Citrix, Notes, ■ The Stack Graph performance utilization and measuring tool. Exchange, WebSphere, Oracle, Polyserve, Microsoft, and others. This tool can monitor Ethernet, processor, storage utilization, and provides a graphical representation to the system adminis- On the systems side, Intel and IBM announced in September a trator. You can imagine this facility integrated with the collaborative effort where the two would be involved with devel- dynamic re-provisioning aspect of Control Tower can result in oping and expanding the BladeCenter architecture. Specifics on a very powerful automated provisioning environment. this program aren't available at the time of this article. RLX uses an interesting mix of processors in their compute blade Rob Sauerwalt with IBM commented, “The IBM BladeCenter offerings. The first blades offered by RLX used the Transmeta gives customers the opportunity to reduce complexity and expand processor, compatible processors designed for low power service offerings, all at a lower total cost of ownership.” For more and reliability. Transmeta-based blades from RLX range from information, visit www.pc..com CPU633 at 633 MHz to the 1000T at 1 GHz. Last year, RLX pro- duced two new Intel-based server blades: the 800 MHz Pentium RLX III based 800i and most recently the 1.2 GHz 1200i designed The RLX solution is designed for large scale-out enterprise with the Intel Xeon processor. The latest blades support dual deployments, provides advanced management and provisioning Xeon processors. software, and can pre-install mix and match integrated application software with their systems. At the time of this article, the RLX blade server solution is 10/100-class I/O. By the time this article is released, RLX expects RLX was founded in May of 2001 with the single focus of devel- to have a new blade server solution providing Gigabit Ethernet oping blade server solutions for the enterprise market. RLX I/O. In addition, RLX is exploring other low latency, high-speed recently closed $15M in venture capital bringing the total invest- interconnects such as Infiniband. ment in the company to $74M. At a time when venture capital investments are few, RLX continues to be a very credible player When it comes to software partners, the usual cast of characters and pioneer in blade server technology. including Red Hat and Microsoft were mentioned. Since the com- plete focus is server blades, RLX had to tackle putting Windows RLX blade server enclosures support up to 24 servers with a pair images on blade servers in headless operation (no keyboard, mon- of hot-pluggable power supplies being shared by 24 servers, each itor, or mouse). Due in part to RLX, Windows 2003 is announced with a 10/100 Ethernet connection. The goal of the system is to to have better support for headless operation. consolidate and shrink cabling over stacked PC servers by up to 8:1 reduction. There are three 10/100 Ethernet interfaces on every RLX also integrates specific applications in their systems for var- blade Ð two for I/O and one dedicated for management. It’s ious customers. One interesting and unconventional application through the management port that communication with the RLX was BLAST, an application used in the bio-technology industry Control Tower management software occurs. for doing genomic searches. BLAST is an open source applica- tion and runs effectively across multiple Intel-based nodes in The Control Tower management system is a significant feature of scale-out server blade environment. John Schmitz, product man- the RLX blade server solution and serves as a source of differen- ager at RLX, mentioned RLX is branching out from compute tiation for the company. engine/enclosure/management to a more complete, pre-loaded application specific blade server environment in the future. For Control Tower provides: more information visit www.rlx.com.

■ Basic hardware monitoring, alerting, and normal infrastructure

Reprinted from CompactPCI Systems / March 2003 Sun cited specific standard groups or organizations as important Sun’s main area of distinction is the incorporation of a heteroge- to the advancement of blade server technology as PICMG and the neous blade server approach. This distinction is evidenced in their Service Availability Forum. PICMG is an important form factor organization, layered network management system, and blade standards body and unifying network management function for server accommodation for x86 and Sparc with Linux and Solaris the industry. operating systems. The Service Availability Forum is an industry group comprised of Two organizations within the Sun Microsystems Volume Systems Sun, Intel, Cisco, GoAhead, and others who are pioneering HA Products division are the telecom oriented division with Sun Netra software from operating system to application. These efforts products and the enterprise division focused on enterprise blade include Windows, Linux, and Solaris operating systems. server solutions. The organization within Sun working on the Enterprise Blade The Sun Netra solution is best described as a communications product has published white papers on the subject, but at the time market product evolution Ð the use of blade form factors has been this article is being written, no products have been announced. around in communications for a long while. However, the Netra Ashley Eikenberry with the Sun enterprise blade server group products mix CompactPCI form factor with a blade server men- mentioned that a the heterogeneous platform with a layered man- tality, enabling a wider variety of I/O to be accommodated. The agement solution at the blade, chassis, and rack level that's man- latest in the Sun Netra architecture products are the CT410 and aged through common tools is key to the proliferation of blade CT810. Sun has taken availability scalability one step further with servers for a variety of applications. The enterprise organization the concept of a chassis being further divided into drawers. Sun of Sun is looking at blade server applications in a wider range of has an architecture focus on high availability, incorporating things markets ranging from financial services to billing systems and like satellite I/O slots that can share support resources such as technical computing. Ashley mentioned Sun is not just developing storage, fans, and power supplies. High-availability enhancements a product line Ð they are also focused on working with prospects like HA protocols developed for use in conjunction with TCP and customers to innovate how blade server technology can be optimize performance as well as providing a higher level of reli- applied to address challenges in a variety of markets. For more able communication between blades. information visit www.sun.com/servers/entry/blade/.

The latest Netra Telecom compute blade uses an UltraSPARC IIi 650 MHz processor with Solaris 8 OE or later. The CT 410 houses 16 blades per shelf and 48 processor blades per rack. The external I/O for the satellite blades is 10/100 standard with PMC options. There is IPMI communication for lowest common denominator management functions. The roadmap includes moving the data rates up a level with Gigabit and I/O and between chassis.

Sun mentioned they value partnerships. Application specific blade deployment is so fragmented, it's difficult to offer applica- tion specific blades as a standard part of their product line. Arlen Vanderwel commented, “For example, if we offered a specific partner SS7 blade as a standard product add-on, we'd risk losing a large percentage of our customer base. Every customer likes his own application specific blade vendor. Sun accommodates by making sure our system management software is heterogeneous across any CompactPCI-based board vendor.” Sun has also formed a Customer Ready Services group within Sun. This group provides integration services that provide a mix of the Sun blade server products along with application specific blades delivered in turn-key fashion.

Reprinted from CompactPCI Systems / March 2003